DSpace Repository :: Browsing by Author "Atkison, Travis"

Browsing by Author "Atkison, Travis"

Now showing 1 - 11 of 11

Autonomous Vehicle Handling Analysis for an Indy Autonomous Challenge Car
(University of Alabama Libraries, 2023) Frederick, Robert Cole; Dixon, Brandon D
This dissertation aims to present a handling analysis package developed to keep the car from crossing the limit of traction for the first fully autonomous head-to-head race. The Indy Autonomous Challenge is a fully autonomous racing competition with ten university-based teams competing with identical cars. The University of Alabama participated in the IAC as a part of the PoliMOVE team. The competition challenged teams to integrate software and hardware and develop algorithms for autonomy and control. A portion of the software and hardware integration is the vehicle's control and dynamics, which is essential to keep the car on track. This research focuses on determining methods and algorithms to analyze the vehicle's handling to keep the car below the traction limit. An understeer/oversteer controller is proposed to solve the negative effects from the controller on a lateral controller. An application was created to analyze data and to rapidly develop algorithms for interrupting the raw vehicle data. Quick data analysis is critical to managing the vehicle's handling, which is why the team analyzes data live through telemetry. In contrast, other data is post-processed to ensure the car's safe operation. Crashes cost tens of thousands of dollars and can cause the team to miss crucial test sessions for developing the vehicle. The paper presents the analysis package, controller, and algorithms designed for the PoliMOVE team car.
Enabling Incremental Federated Learning for Autonomous Driving: a Network Perspective
(University of Alabama Libraries, 2022) Subedi, Pawan; Hong, Xiaoyan; University of Alabama Tuscaloosa
Autonomous driving relies greatly on deep learning models to comprehend the surroundings and activities of the road systems. These learning models are traditionally trained off-line and used during driving. However, recent research on federated learning has enabled distributed deep learning for model adaptation with new data inputs from end users. Similarly, the recent research on incremental learning has enabled the upgrading of the learned model with newer rounds of training, without losing previously acquired knowledge. For autonomous and/or connected vehicles (AVs, CVs), these mean it is possible to take new, maybe real time, data inputs from various sensors from multiple vehicles in vicinity, to train and update the preloaded models. The updated models can improve the safety and reduce human involvements when driving through unfamiliar situations. Despite the tremendous possibilities of federated learning for autonomous driving, it faces several challenges for its implementation. The challenges include the heterogeneity of the end devices, statistical heterogeneity of the available training data, emerging privacy challenges, energy management, model data distribution, etc. Although recent works in the literature attempt to answer these challenges, the assumption of identical distribution of the training data in most of these solutions is difficult to achieve. Similarly, all these works lack the consideration of the challenges for distribution of the model for training in dynamic and mobile edge wireless environment.To this end, this thesis presents a networked system using Fog and Edge architecture utilizing VANET and VDTN for model training management answering the need for distribution of the model. A novel scheme for federated learning utilizing the networked systems is then presented. The presented scheme for federated learning has quicker convergence and also considers the presence of data with varied distribution at the end devices. Finally, to answer the need for reliable data dissemination in case of lack of VANET and VDTN based communication, a complimentary network system utilizing the ultra dense network is presented.
Mining and Ranking Incidents for High Priority Intrusion Analysis
(University of Alabama Libraries, 2020) Haque, Md Shariful; Atkison, Travis; University of Alabama Tuscaloosa
Threats and intrusions are increasing at an alarming rate, even though related technologies have observed rapid advancement. Hence, advanced threat analysis has become imperative to improve current technologies. These technologies are primarily designed to detect or predict threats and minimize the likelihood of damage. The goal of an efficient intrusion analysis is also to develop models unwavering to any external influences and produce optimized results. Several data mining techniques have been applied in these scenarios to detect both anomaly and misuse, predict possible attack paths, or generate attack models. Some consider determining the priority, an important criterion of alerts, using different characteristics of the attack scenarios. In this dissertation, novel priority-based alert mining techniques and a ranking model are proposed to prioritize sequences of alerts and to realize their actual effect which is often misunderstood due to the generic taxonomies used by detection systems. This dissertation has the following contributions: First, a novel data mining-based alert sequence mining technique is proposed to discover potential attacks from intrusion alerts. Intrusion detection systems maintain signatures of intrusions with a severity scale. This information has been leveraged predominantly in the proposed data mining-based alert association approach. This approach reduces the effort of post-processing alert sequences and calculating their severity when the relationship is established. Second, a non-redundant high priority association rules mining technique is proposed based on theories and background of non-redundant association rule mining. Such techniques are highly adopted to determine the correlation between items in sequences and to develop efficient prediction models with a reduced volume of derived data. Third, the above mining approaches facilitate the process of extracting severe incidents based on priority. However, severity levels determined by the detection system are generic; thus, their real consequences are hard to perceive. Multi-criteria decision making is a prominent research area to assess different alternatives. The proposed approach is equipped with a combination of MCDM techniques to further rank the prioritized threats based on several benchmarks. The novelty of our technique is to consider the priority level of alerts at prior stages of attack analysis and later determine the overall attack scenario.
A novel intersection-based clustering scheme for VANET
(University of Alabama Libraries, 2021) Lee, Michael Sutton; Atkison, Travis; University of Alabama Tuscaloosa
Currently, much attention is being placed on the development and deployment of vehicle communication technologies. Such technologies could revolutionize both navigation and entertainment systems available to drivers. However, there are still many challenges posed by this field that are in need of further investigation. One of these is the limitations on the throughput of networks created by vehicular devices. As such, it is necessary to resolve some of these network throughput issues so that vehicle communication technologies can increase the amount of information they exchange. One scheme to improve network throughput involves dividing the vehicles into subgroups called clusters. Many such clustering algorithms have been proposed, but none have yet been determined to be optimal. This dissertation puts forth a new passive clustering approach that has the key advantage of a significantly reduced overhead. The reduced overhead of passive algorithms increases the amount of the network available in which normal data transmissions can occur. The drawback to passive algorithms is their unreliable knowledge of the network which can cause them to struggle to successfully perform cluster maintenance activities. Clusters created by passive algorithms, therefore, tend to be shorter-lived and smaller than what an active clustering algorithm can maintain. In order to maintain a cluster with a low overhead and better knowledge of the network, this dissertation introduces a new clustering algorithm intended to function at intersections. This new algorithm attempts to take advantage of the decreased overhead of passive clustering algorithms while introducing a lightweight machine learning algorithm that will assist with cluster selection.
Particle Swarm Based Reinforcement Learning for Path Planning and Traffic Congestion
(University of Alabama Libraries, 2022) Phan, Ashley; Atkison, Travis; University of Alabama Tuscaloosa
In 2019, the average American commuter wasted approximately two and a half days due to traffic delays. Researchers suggest that these delays could be relieved by the addition of intelligent transportation systems, such as navigational systems that identify multiple high-speed travel routes or sophisticated traffic signals that can adapt to different traffic patterns. This dissertation explores the hybridization of the swarm intelligence algorithm, particle swarm optimization, with the reinforcement learning algorithm, Q-learning, and the hierarchical reinforcement learning algorithm,MAX-Q, to produce an intelligent path-planning algorithm and an adaptive traffic control system. By combining these algorithms with particle swarm optimization, the search space of a single agent is reduced through the parallelization and collaboration of multiple agents. Alternatively, the use of a look-up table improves the performance of particle swarm optimization by enhancing the swarm's ability to learn and balance the local and global search. In order to further improve the performance of the hybrid algorithms, a local particle swarm optimization variant was incorporated into the algorithms' action selection policies. This combination results in two hybrid intelligent optimization algorithms, Q-learning with Local Particle Swarm Optimization and MAXQ with Particle Swarm Optimization. When tasked with path planning in the Taxi World environment, QLPSO and MAXQPSO collectively learned the optimal policy in 46.44% fewer episodes than state-of-the-art algorithms and completed the task in 25.57% fewer steps. Given the success of the novel methods in the path planning problem, the two algorithms were slightly modified to identify the optimal policies for the traffic control problem. For various traffic networks, the algorithms collectively minimized the total wait time by an average of 16.31% and decreased the average wait time per vehicle by 11.43%. The combination of PSO and the learning algorithms demonstrate notable benefits as intelligent transportation systems.
A Predictive Approach to Detect and Mitigate Sybil Attacks in a Waiting Time-Based Adaptive Traffic Signal Controller
(University of Alabama Libraries, 2024) Al Ani, Sinan Ameen; Atkison, Travis
An adaptive traffic signal controller (ATSC) uses the trajectory data transmitted from the vehicles to maintain the green time of a signalized intersection in connected transportation systems. Existing studies show that ATSC can significantly decrease intersection waiting time, thus improving travel times along signalized corridors. However, an attacker can generate fake vehicles and corresponding basic safety messages to cause traffic congestion in a connected vehicle (CV) based ATSCs. This form of attack, known as the Sybil attack, can create traffic congestion by injecting fake vehicles slowly approaching the subject intersection. Hence, it alters the signal timing and phase without causing abrupt changes in the number of vehicles. This dissertation introduces a predictive approach to detect and mitigate Sybil attacks for a waiting time-based ATSC algorithm utilizing a type of recurrent neural network known as a long short-term memory (LSTM) model. A microscopic traffic simulator--Simulation of Urban Mobility (SUMO) was utilized to generate this type of attack. Furthermore, data extracted from the simulator containing vehicle waiting time and number of vehicles was used to develop the prediction model. The detection approach was evaluated on three different datasets with sizes of 500,000, 2,000,000, and 5,000,000 observations. The results showed that the detection approach when trained on the 500,000 dataset can effectively detect Sybil attack, achieving an accuracy of 91% on under attack data. Building on this robust detection framework, a mitigation strategy has also been developed. In order to mitigate this attack effectively, it is essential to predict the number of vehicles traveling from each attack-free adjacent intersection to the subject intersection. The results of the mitigation strategy demonstrated that the approach effectively predicts the flow of vehicles that come from the adjacent intersection to the subject intersection.
Redefining privacy: case study of smart health applications
(University of Alabama Libraries, 2019) Al-Zyoud, Mahran; Carver, Jeffrey; University of Alabama Tuscaloosa
Smart health utilizes the unique capabilities of smart devices to improve healthcare. The smart devices continuously collect and transfer large amounts of useful data about the users' health. As data collection and sharing are two inevitable norms in this connected world, concerns have also been growing about the privacy of health information. Any mismatch between what the user really wants to share and what the devices share could either cause a privacy breach or limit a beneficial service. Understanding what influences information sharing can help resolve mismatches and brings protection and benefits to all stakeholders. The primary goal of this dissertation is to better understand the variability of privacy perceptions among different individuals and reflect this understanding into smart health applications. Towards this goal, this dissertation presents three studies. The first study is a systematic literature review conducted to identify the reported privacy concerns and the suggested solutions and to examine whether the context is part of any effort to describe a concern or form a solution. The study reveals 7 categories of privacy concerns and 5 categories of privacy solutions. I present a mapping between these major concerns and solutions to highlight areas in need of additional research. The results also revealed that there is a lack of both user-centric and context-aware solutions. The second study further empirically investigates the role of context and culture on the sharing decision. It describes a multicultural survey and another cross-cultural survey. The results support the intuitive view of how variable privacy perception is among different users and how understanding a user's culture could play a role in offering a smarter, dynamic set of privacy settings that reflects his privacy needs. Finally, the third study aims at providing a solution that helps users configure their privacy settings. The solution utilizes machine learning to predict the most suitable configuration for the user. As a proof of concept, I implemented and evaluated a prototype of a recommender system. Usage of such recommender systems helps make changing privacy settings less burden in addition to better reflecting the true privacy preferences of users.
Simulation and Analysis of Traffic Congestion Prediction and Vehicle Re-Routing Strategy Using Image-Based Surveillance Camera
(University of Alabama Libraries, 2023) Wang, Chen; Atkison, Travis
Traffic flow management is crucial for intelligent transportation systems, as congestion in arterial areas, highways, during bad weather, and rush hours is increasingly prevalent. Efficient traffic flow detection, prediction, vehicle re-routing, and active travel planning are essential for transportation system management. However, upgrading hardware infrastructure like radar, cameras, and detection tools to keep up with evolving vehicle tracking algorithms is costly and time-consuming. This dissertation reviews different approaches to vehicle tracking, short-term congestion prediction, and mitigation. Based on previous research, a cost-effective integrated congestion awareness system called the heat-balancing path planning system is proposed. The system predicts and detects congestion, balances traffic flow, and reduces overall congestion potential. It comprises the same vehicle recognition, short-term congestion detection and prediction, and passive vehicle notification with dynamic re-routing. Leveraging existing traffic surveillance cameras, this method offers a viable solution for regions without additional hardware investments. The core methodology of the proposed system is inspired by thermal-transfer characteristics in materials. The model predicts congestion based on vehicle volume heat-density and traffic speed. Simulated annealing is used to suggest a traffic-balancing plan, and a dynamic re-routing algorithm is adapted from the k-shortest path algorithm. The model is implemented in a custom-designed simulated traffic flow environment, mimicking real-life conditions. Simulation tests validate the model's performance, preventing 28.75\% of congestion, suppressing 63.39% of congestion within a minute, and increasing average travel speed to 72.92--76.15% of the speed limit. Compared to other approaches, the proposed method consistently outperforms, reducing travel time and maintaining higher average speeds. This advantage, combined with practical implications for transportation management, makes it a promising solution for modern traffic challenges. In summary, this dissertation introduces a cost-effective, integrated method for traffic flow management, featuring novel heat-balancing path plan algorithms. Simulation and comparisons with existing methods demonstrate their merits. This work has the potential to enhance transportation system efficiency, especially in regions with limited infrastructure resources.
Transportation Digital Twin Framework and its Vulnerabilities Against Cyber-Attacks
(University of Alabama Libraries, 2023) Irfan, Muhammad Sami; Rahman, Mizanur
Digital twin (DT) systems aim to create virtual replicas of physical objects that are updated in real-time with their physical counterparts and evolve alongside the physical assets throughout their lifecycle. Transportation systems are poised to significantly benefit from this new paradigm. In particular, DT technology can augment the capabilities of intelligent transportation systems. However, the development and deployment of networkwide transportation DT systems need to take into consideration the scale and dynamic nature of future connected and automated transportation systems. Therefore, there is a need to understand the requirements and challenges involved in developing and implementing such systems.This thesis investigates the development of a Transportation DT (TDT) system framework and all related components alongside an assessment of the vulnerabilities of such a system against cyber-attacks. Accordingly, the concept of DT and its relationship with the transportation system is investigated. Current studies on the safety and mobility enhancement applications using DT are surveyed. A hierarchical concept for a TDT system starting from individual transportation assets and building up to the entire networkwide TDT is presented. A reference architecture is also presented for TDT systems that could be used as a guide in developing TDT systems at any scale within the presented hierarchical concept. The study also investigates the vulnerabilities of the system against cyber-attacks with respect to each part of the architecture. Based on the vulnerability assessment of the system, an intelligent attack model is developed that uses a Reinforcement Learning (RL) based attack agent to execute a sybil attack on the system. This attack model is implemented in a simulation scenario using a microscopic traffic simulation software with the goal of creating congestion within a TDT system. The analyses revealed that the RL agent can learn an optimal policy for creating an intelligent attack.
Ultra-Wideband Monocone Antenna and V2X Testing
(University of Alabama Libraries, 2022) Lee, Wooseop; Jeong, Nathan; University of Alabama Tuscaloosa
In this thesis, a capacitively-fed, ultrawide bandwidth, low profile, Omni-directional monocone antenna is proposed for V2X wireless communications. The proposed antenna consists of five main components – circular monocone, capacitive feed, grounded ring, ground post near capacitive feed, the short and long meander grounding vias. The proposed antenna is modeled with an electromagnetic simulator and validated with measurement. The results show that the proposed antenna supports ultrawide bandwidth from 0.75 GHz to 7.47 GHz mounted on a ground plane, allowing GSM, CDMA, UMTS,LTE, GPS, WiFi, BT, DSRC, and C-V2X bands. The prototype of the antenna is 3D printed with low-cost plastic material and sprayed with copper particles for rapid and cost-effective fabrication. The diameter and height of the antenna are 148 mm and 26.695 mm, respectively. The efficiency is measured to be over 87.97 % throughout the frequency bands of interest. Proceedings from antenna design and measurement, different cases of V2X testing were conducted. The Line of Sight (LOS), Non-Line of Sight (NLOS), Intersection, and three different shadowing tests were examined and evaluated with Cohda Wireless’s DSRC and C-V2X supportive units.
Wineinformatics: Using the Full Power of the Computational Wine Wheel to Understand 21st Century Bordeaux Wines from the Reviews
(MDPI, 2021) Dong, Zeqing; Atkison, Travis; Chen, Bernard; University of Central Arkansas; University of Alabama Tuscaloosa
Although wine has been produced for several thousands of years, the ancient beverage has remained popular and even more affordable in modern times. Among all wine making regions, Bordeaux, France is probably one of the most prestigious wine areas in history. Since hundreds of wines are produced from Bordeaux each year, humans are not likely to be able to examine all wines across multiple vintages to define the characteristics of outstanding 21st century Bordeaux wines. Wineinformatics is a newly proposed data science research with an application domain in wine to process a large amount of wine data through the computer. The goal of this paper is to build a high-quality computational model on wine reviews processed by the full power of the Computational Wine Wheel to understand 21st century Bordeaux wines. On top of 985 binary-attributes generated from the Computational Wine Wheel in our previous research, we try to add additional attributes by utilizing a CATEGORY and SUBCATEGORY for an additional 14 and 34 continuous-attributes to be included in the All Bordeaux (14,349 wine) and the 1855 Bordeaux datasets (1359 wines). We believe successfully merging the original binary-attributes and the new continuous-attributes can provide more insights for Naive Bayes and Supported Vector Machine (SVM) to build the model for a wine grade category prediction. The experimental results suggest that, for the All Bordeaux dataset, with the additional 14 attributes retrieved from CATEGORY, the Naive Bayes classification algorithm was able to outperform the existing research results by increasing accuracy by 2.15%, precision by 8.72%, and the F-score by 1.48%. For the 1855 Bordeaux dataset, with the additional attributes retrieved from the CATEGORY and SUBCATEGORY, the SVM classification algorithm was able to outperform the existing research results by increasing accuracy by 5%, precision by 2.85%, recall by 5.56%, and the F-score by 4.07%. The improvements demonstrated in the research show that attributes retrieved from the CATEGORY and SUBCATEGORY has the power to provide more information to classifiers for superior model generation. The model build in this research can better distinguish outstanding and class 21st century Bordeaux wines. This paper provides new directions in Wineinformatics for technical research in data science, such as regression, multi-target, classification and domain specific research, including wine region terroir analysis, wine quality prediction, and weather impact examination.

Browsing by Author "Atkison, Travis"

Results Per Page

Sort Options